Are the MOJ figures that Andrew Doyle a sample or are they a population? That is, are those figures based on a sample(s) of prisons and prisoners, or are they a direct count of all the (self-styled) trans prisoners in UK (or do I mean English?) prisons? Because if this is a count of all of them, even if they're a small number, that doesn't make them unrepresentative. They can't "represent" anything, they're not a sample of the population of trans prisoners they are the whole population of "transwomen in prison" who need to be risk assessed. And if the proportion of self-styled transwomen in prison who are sex offenders is higher than the proportion of other prisoners or other male prisoners who are sex offenders then it's not a sampling error it's just a fact. A plain, ugly, hard to swallow fact.
As for "prisoners must be assessed individually" well they aren't now, they never have been, and they never will be. Risk assessments are based on group characteristics, like is a prisoner male or female; a known sex offender or not; violent or not; etc. This idea that prisons do some delicate identification of who poses a risk based on profound psychological insights (that prisons have and the rest of us don't) is somewhere between naive and dishonest.
It is true that the MOJ figures don't say anything about transwomen in general, outside of prison. Prisoners are not a representative sample of the population who are not in prison, trans or otherwise.
But those MOJ figures should affect a prison risk assessment. That's how risk assessment works.